When porting the gitpod-io/memo blueprint to a production repository, why must the engineering team construct the harness (AGENTS.md, docs/architecture.md, and verification scripts) before letting agents modify product code?
Build Your Own Factory on a Real Repository
Take everything from the course and stand up your own software factory on a repository of your choice, mirroring the memo blueprint. Choose the product, build the harness, wire the four stations, add triggers and evals, then raise autonomy only when the evidence supports it.
Loading lesson content...
An engineer configures the Station 3 verifier subagent manifest shown below. Why must the verifier possess shell execution permissions while having edit permissions strictly denied, and operate in an isolated subagent session?
A team operating a Level 3 software factory proposes removing human review gates at Station 4 to enable autonomous Level 4 merges for dependency updates. Over the last month, their trailer ledger records 14 runs with a 78% FPVR, an average review duration of 9 minutes, and two flaky integration tests. What should the engineering lead decide?